Separating Tones from Noise.

Yes. The easiest way I can think of is with multiple notch (or narrow band pass)filters.

http://en.wikipedia.org/wiki/Band-stop_filter

Depending on the application it can be done in either the analog or digital domain (digital domain using digital signal processing is more relevant to your application I think, but it is more complex). For general information and circuits try a google search of this term.

For more specific answers you will need to supply more information on your application including information on the tones, noise and accuracy and signal attenuation needs, tone duration and required speed of filtering method, number of tones needed to be filtered, etc.

Awhile back what I did trying to get the best sound possible from cassette tape recordings was to push all through a Realistic equalizer and a Panasonic Reverb into a Teac CD stand alone Recorder.

I bought a Sony Mini Disc recorder to make better recordings in the first place.

For awhile I did have a NAD amp that helped quite a lot, for final monitored laydowns.

CD masters now are all that play. Copies from masters are not possible now. Sony Mini Disc broken. NAD, broken.

Room Tone. cut that out.

A book by the recordists of Steely Dan would be good.

Not true due to Tinitus, or ringing in the ears.

I don't mind much going deaf, but I do mind hearing about it.

"an electronic circuit to separate audio frequency tones from noises"

They are in all the "noise cancelling" headsets on the market today. These headsets are pretty good at filtering out repetitive noises and some dirty noise, as well. They rely on a sampling microphone, a cancellation circuit and a transducer to produce the out of phase waves to cancel out the offending noise.

The best way to separate fixed frequencies from the noise is to use synchronous demodulation....

This form of demodulation allows you to dig out the signal even when its buried in noise....As the input is demodulated at the same frequency as the signal the output is directly related to the intensity of the signal (and the phase if you need that information).

The even order harmonics will get through though...

John.

A phase lock loop, acronym PLL, is ideal for finding a signal buried in noise. It works on the principal that the noise is random and the wave is recurring.

These are made in IC form with lots of data sheets on line, as this search shows.

http://www.google.ca/#hl=en&source=hp&q=%22phase+lock+loop%22&btnG=Google+Search&meta=&aq=f&oq=%22phase+lock+loop%22&fp=f0935e104527101b

No, phase lock loop alone will not do as one has to get the signal from a composite source "noise". Box car averager may be a better idea or simply a spectrum analyzer with scanned mixer frequencies to extract different frequency sources that get tunned like AM Radio.

Voice recognition chips some time used multiple band filters or scanned band filters. I have used one such 16-channel version perhaps was SA-16. It is also possible to use digitally programmable analog filters of Maxim Inc. USA.

I beg your pardon, a PLL is exactly what is needed to find a sine wave in a sea of noise.

You can find a sine wave with a PLL where a spectrum analyzer shows a sea of 'grass'.

I suggest you look at front end receiver design issues.

http://www.google.ca/#hl=en&source=hp&q=%22radio+receiver%22+%2B%22front+end%22+%2BPLL&btnG=Google+Search&meta=&aq=f&oq=%22radio+receiver%22+%2B%22front+end%22+%2BPLL&fp=f0935e104527101b

You don't specify whether you have one or a few specific frequencies, or whether you want a major part or the whole audio spectrum. If you have only one or a few specific frequencies of interest, then it's easy with a resonant circuit for each desired frequency. the resonant circuit can be RC, LC (better), or a high Q active filter.

Yes, since noise is normally distributed and the tones are not, you can use a Fourier transform to map the signal into the frequency domain where the tones will stand out. (some details are missing)

This is often done in astronomy and telecommunications so the mathematics and associated real world algorithms & circuitry (usually using DSP's) is covered in most elec engineering texts.

Be warned, it's pretty heavy going even if you do have a background in maths.

Take a long look at Dolby drive-rack, this is it's (compression) essential function.

Several audio recording softwares are capable also to isolate a myriad variety of frequencies from the desired signal path.

All depends on if you need to take out the noise "inline", or you want to take the noise out of some kind of recording. Also very important is to know how much noise is on the signal. Do you want to "dig out" a signal out of a lot of noise, or do you want to "cleanup" a noisy signal.

1) "INLINE"
a. use an equaliser (preferable a parametric one) to filter out the frequenties you are interested in. (dig for signal)
b. use a "noise gate" (sometimes called a "squelch"). This is an amplifier that does not amplify when the input signal is below a certain level. (cleanup)
c. use a DNL (Dynamic Noise Limiter) This circuit narrows the bandwidth depending the input signal : High input > all frequencies pass, Low input : bandwidth is limited (cleanup)
d. A combination of a + b or a + c or b + c or a + b + c. Remark : depending what kinda noise, it might be worth trying to change the order of "effects".

2) An existing (recorded) signal.
Of course all the effects from "inline" can be used. If you have the recording in didital form, you can use a program that has all those effects on board. Most of these programs also have some kinda special noise reduction system. You can take a sample of the noise (when there is NO important signal) and substract this (more or less) from the whole recording. Usually this works very good !

Good luck and Have Fun !

I think this could be easier than some proposed solutions suggest. You wrote "where the two are generated serially". If "audio frequency tones" is interpreted to mean sine waves, and if the frequency of those tones is known, then this is very simple. In that case, you only have to discriminate between a sine wave of a particular frequency (and to some extent its multiples) when the two occur one after the other.

From the way your question is written, the challenge is not to dig particular "sounds" out of noise, but to tell when a sound is very close to a single sine wave vs a sound that has the random frequencies we think of as noise, one after the other.

I think you may need to clarify exactly what you want to do.

If you want to separate "music" from "noise" when the two occur simultaneously, then you have a very challenging project: just ask any old person... most of today's music is just noise.

If you want to separate "music" from "noise" when the two occur simultaneously, then you have a very challenging project: just ask any old person... most of today's music is just noise.

I find separating unwanted frequencies from the desired less than awkward.

Notch filters are fine if the disruptive frequencies are known. I suppose I'm just lazy and haven't the will to hunt down offending frequencies preferring to employ a feedback killer instead

Older folks with less flexibility of the membranes (eardrum) is not unlike trying to capture rapid pulses with a heavy moving coil device. The sound is similar to several voices inside a barrel; gross inaccuracy.

You may find this thread very interesting: http://cr4.globalspec.com/thread/43628#newcomments

Sounds like how a 2-tone sequencial pager works. Circuitry past the disciminator is what would be needed.

Why reinvent what is already built?

Trying to improve it may invent new uses...

What do you mean by tones, one, two, several are they constant in frequency? And how are those tones produced (in a radio or so?)

As a telegraphist I've experimented a lot with it and believe me you can do miracles those days.

JP

Dear on7ami: The 'tones' are the audio frequencies of the human voice frequencies, roughly say 60 to 3500 hertz. The tones that I am concerned with are those that are voice generated which we call 'vowels'. The 'noises' that I am trying to separate from these tones are 'noises' that we are mostl familiar with, they are the 'consonants' of speech. Even though they are considered 'noises' I would prefer to call them 'characteristic noises' because we recogize them readily during ordinary apeech. In fact consonants are the very basis of understanding speech. For example, the word 'cat' without the 'noises' of K and T has no meaning. ......Finally, why do we want to separate 'tones' from 'noises'? The answer is that vowels can be extremly loud by comparison with consonants. The vowels are generated starting with vibrating the vocal chords and can these vibrations can then be amplified to extreme levels by use of the cavitities in the skull. Not so for the consonants which are formed through manipulation of the mouth, lips and tongue. There is essentially NO amplification of the consonants. As a result of the mismatch of sound levels, millions of hard of hearing (HOH) folks waste their money buying expensive hearing aids that amplify all audio frequencies. Leaving us HOH people crying out "I hear but I don't understand". We hear the loud vowels but not the feeble consonants, hence 'I don't understand'. I can provide aditional information as needed....... There should be tremendous interest in doing this 'amplification of consonants' because the economic benefits will be big, big, big. If you can do some 'miracles' for the world of voice communication it will pay handsomely.

Al Mond,

As a radio amateur, we had an alike problem when transmitting in single side band. The sound generated by the vowels, generally determine the maximal power of the transmitter and we would like to use the power on an optimal way so that all information is send to the other side with a same level of energy. Separate vowels and consonants, I think, will be a huge challenge. But we use what we call audio-compressors, devices who regulate their gain with the level of the signal. So vowels are less amplified then other noises. That works well as long as you speaks, the moment you stop speaking, background noise is amplified strong too so you have to build in some limits etc.

One can realize this solution on an analog way or on a digital way using DSP. If is just for speech you can limit the bandwidth to.

If you go to look on the site of microchip in de developers part under audio processing, you can find already good information.

If it helps, give a cry, I have some other info too.

Greets

Jean Paul

Why didn't you say that in the first place?

Very interesting concept! Fricatives and stops (F and T, for example) generally involve much higher frequencies than vowels, so I'd think that an amplifier with higher gain at higher frequencies (up to some limit) should be a good start.

Surely the better hearing aids already do that... (I'm assuming)?

I'm in the 'hear but don't understand' group myself, but so far have avoided trying hearing aids. One of these days I need to do that...

Good luck on your project!

Dear dkwarner: The first to make us aware of a hearing problem is typically one's spouse. After months and years telling her she mumbles, you finally come to the realization that the fault is yours, not your spouse's, not the rest of the mush-mouthed world. OK. Now you go out and buy a set of in-the-ear hearing aids. Big money outlay. Can't stand the new sound with the hearing aids. It sounds like you are inside a confined barrel. OK. Dump the aids in the dresser drawer. Now that we poorer and wiser so this time we go out and buy a new set of hearing aids, a behind the ear model. Ah much better, no barrel sounding audio. Next problem. Can't distinguish between words like moon, soon, loon, coon, tune from one another. In other words, can't hear/distinguish the consonants. That's where it gets rough. Got to do something with those weak consonants. They are so infinitly weaker than the vowels. This is an important aside. The consonants are not as typically verbalized as being, Beee. Ceee, and Dee, etc. The consonants are in reality just the B,C,and D without the assisting vowels. Now you can begin to see that when these consonants are spoken its as if you were whispering so much so that it is necessary to add a vowel following the consonant to make an audio sound loud enough to be heard. ..... I've tried to be tuned into as many individuals and groups as possible and I've discussed the problem with many hard of hearing (HOH) folks as I can. They seem to readily understand the problem and encourage me to go ahead and quickly patent and build a working model....Couple of problems with that, starting with my age-89 y/o, no design experience, no lab. Excuses, Excuses, Excuses. ....However, my lifelong experience in the Radar field makes me readily aware that this is a new kind of modulation problem, I'll give it a name- Serial Modulation. Why Serial? Because vowels and consonants follow one another sequentially. Example, the Kaa sound, followed by the aah sound, and followed by the Tea sound creates the sound that says 'cat'.

We HOH souls turn up the radio and TV audio to excess (to the normal hearing pearing person) simply because we want to hear those weak little consonants. Turn up the volume on only the consonants and we won't have need for the excesive loudness as we presently do. I view it as a go, no-go situation. If its a vowel turn down the volume, if its a consonant take advantage of the full gain of the amplifier, it won't break anyone's ear drum. Computer logic can handle the go, no-go situation readily. Vowels can be readily distinguished from consonants. Vowels consist of frequencies that are repetative in nature. Not so the consonants. Is that sufficient to separate the two. I think so. regards, Al

As a teenager, I used guns, dynamite, and chainsaws (all while doing paid jobs!) - all ignorantly without any form of hearing protection. As a result, I have a known notch at 8khz, tinnitus whenever I stop to listen, and that characteristic 'hear but don't understand' problem.

Yes, my wife has been telling me to get hearing aids for some time now...

Back in the '70s, I actually designed and built a hearing test unit (using 4000 series CMOS ICs). Unfortunately, there was a tiny 'click' between test tones, so the guy I was building it for didn't accept it and abandoned the project. It's still out in the garage... Unfortunately I know almost nothing about DSP (Digital Signal Processing), but it does sound like DSP should be able to do what you are describing. Hopefully now that you have disclosed what you are trying to do (and a noble effort it IS!), someone who does know will step in and help.

Good luck on your project!

Actually, expensive hearing aids do not amplify all frequencies. Some do, to the extent that they have been able thus far, just what you want. Unfortunately, if the little hairs that detect sound have been largely blasted away by loud sounds, I imagine there might not be any practical level of amplification that will restore hearing to original levels.

The characterization of vowels as "tones" and consonants as "noise" is probably hard to support. Desired information in a sound is not noise -- it's signal. What you want is a very high signal to noise ratio, and what you need to do first is define the signal. If that signal is speech, then certain key parts of the information will be at high frequencies. Those frequencies are shared by the frequencies of true noise. For a hearing aid to work well it must amplify the hiss of an "s" but not amplify the hiss of the tires on a car driving by, or the hiss of wind or rain.

You might look into speech to text technology to see how DSP is used. Perhaps syllable recognition is better than vowel-consonant recognition. When you say "a" (aye) the sound is not a "tone" at all. The sound begins with a glottal stop, and then quickly turns into anything but a tone. When you pronounce "father" you do not pronounce "ffff" "aaaa" .... you produce a complicated sound that is recognized as "fa" then another complicated sound that is recognized as "ther". (The aaaa sound would otherwise start with a glottal stop. -- try to pronounce "odd" without the glottal stop, and you end up with something that sounds like "hod")

Also spend some time thinking about singing. The whole range of voice sounds can be reproduced at different frequencies, with an untrained voice being able to reproduce words through a range of two octaves.

M is a consonant, incidentally, that is perhaps much more like a "tone" than any consonant.

Certainly a worthwhile project, and one that many hearing aid manufactures have given a tremendous amount of thought to. When you find the solution, send me one.

I just took a hearing test online. At no point did it provide calibration for my headphones. So there would be no way of knowing if the hearing loss I showed was attributable to my ears of my headphones.

In any case, perhaps the best solution is the one proposed many times by my grandfather: He'd say: "I can hear just fine. The problem is that everyone mumbles." Just get people to stop mumbling.

With modern hearing aids they do a frequency versus sensitivity test for the range you can still detect. Then then use a hearing aid with a built in equalizer to bring your hearing to an acceptable level. Now frequencies you have lost will produce a permanent notch or notches in your perceived range Some can make adjustments to try to compensate for this, but there are limits. These are the high end aid sold by aidiologists. There are a lot of frauds and hucksters in the business, as you can get hearing aid with a lot of gain for about $25 from China, but they lack compensation. If gain is all you need, get the cheap ones. Sadly, the hucksters sell the cheap ones for $1000 or more.

I gather from the discussion that you are looking for some information about filters, and what kind of filter can be applied to reduce the amplitude below 3500 hz where vowels have most of their energy.

You might want to start by readiing the wikis on low pass, high pass and band pass filters. A high pass filter is the type that can be applied to reduce amplitude below 3500 hz, for example.

http://en.wikipedia.org/wiki/High-pass_filter

Audio software like Audacity (or others) has lots of filters available as plug-ins. The better ones allow you to manipulate just how that filter is applied. Cut-off's and notches are good for some applications, but you need some of that energy below 3500 hz for intelligibility. Filters that allow you to manipulate the roll off value, for example, will let you tailor the filter to allow enough in. Bottom line, if the filter doesn't give you much freedom to tailor it, move on and look for another one - there are many high pass filter plug-ins on audacity and they don't all do what you want..

If I were you, I'd start with recordings of the problem sounds, that are good enough you can at minimum identify where the stop and start of words are, at the amplitude that is attainable. Experiment with filters and figure out the filter design that suits you the best.

I could post some images of examples and the filters that worked for me, but my audio work computer is offline at the moment and this one doesn't have the full program. Audacity is nice, it has spectrum analysis in the menu, so you can look at an image of the sound and exactly what you have done to the envelope with your filter. If you're still looking for answers in late november when I have time off, I'll dig up what I've got and maybe (or without a doubt) learn something new myself in this...

cheers.

The PCM (Pulse Code Modulation) used for traditional digital telecomms. is sampled at only 8 KHz, and most telecomm. CODECs have a "brick wall" low pass filter at 3400 Hz. so, I don't believe that there is a lot of very important information in speech above 3500 Hz.

... Just a randomly selected old Texas Instruments CODEC:-

Yep, but the sound quality/intelligibility under the 3400 brick wall isn't great. It leaves out a lot of the energy in consonants like f and s as someone pointed out, which are on the higher end.

Just something I remember from the work I did with noisy recordings: a gently rolling off high pass filter at 600 - 700 hz seemed to enhance intelligibility of the consonant cues more than others.. I think 660 was the 'magic number' for me - but that could well reflect my own hearing imperfections - or the specific noise I was dealing with. That high-pass worked better than filters at 1000 or higher, however.

I'm sure the precise info is out there somewhere, about the frequencies involved in each consonant. One way to check it out would be to make a recording of your own voice with consonant pairs like "da" and "ta", "fa" and "sa". Then look at the spectrum plotted, and see where the differences are.....